1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
50 #include "poll-loop.h"
51 #include "port-array.h"
52 #include "proc-net-compat.h"
56 #include "socket-util.h"
57 #include "stream-ssl.h"
63 #include "vswitchd/vswitch-idl.h"
64 #include "xenserver.h"
66 #include "sflow_api.h"
68 #define THIS_MODULE VLM_bridge
77 /* These members are always valid. */
78 struct port *port; /* Containing port. */
79 size_t port_ifidx; /* Index within containing port. */
80 char *name; /* Host network device name. */
81 tag_type tag; /* Tag associated with this interface. */
82 long long delay_expires; /* Time after which 'enabled' may change. */
84 /* These members are valid only after bridge_reconfigure() causes them to
86 int dp_ifidx; /* Index within kernel datapath. */
87 struct netdev *netdev; /* Network device. */
88 bool enabled; /* May be chosen for flows? */
89 const struct ovsrec_interface *cfg;
92 #define BOND_MASK 0xff
94 int iface_idx; /* Index of assigned iface, or -1 if none. */
95 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
96 tag_type iface_tag; /* Tag associated with iface_idx. */
99 #define MAX_MIRRORS 32
100 typedef uint32_t mirror_mask_t;
101 #define MIRROR_MASK_C(X) UINT32_C(X)
102 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
104 struct bridge *bridge;
108 /* Selection criteria. */
109 struct shash src_ports; /* Name is port name; data is always NULL. */
110 struct shash dst_ports; /* Name is port name; data is always NULL. */
115 struct port *out_port;
119 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
121 struct bridge *bridge;
123 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
124 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
125 * NULL if all VLANs are trunked. */
126 const struct ovsrec_port *cfg;
129 /* An ordinary bridge port has 1 interface.
130 * A bridge port for bonding has at least 2 interfaces. */
131 struct iface **ifaces;
132 size_t n_ifaces, allocated_ifaces;
135 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
136 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
137 tag_type active_iface_tag; /* Tag for bcast flows. */
138 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
139 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
140 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
141 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
142 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
143 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
144 long long int bond_next_rebalance; /* Next rebalancing time. */
146 /* Port mirroring info. */
147 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
148 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
149 bool is_mirror_output_port; /* Does port mirroring send frames here? */
152 #define DP_MAX_PORTS 255
154 struct list node; /* Node in global list of bridges. */
155 char *name; /* User-specified arbitrary name. */
156 struct mac_learning *ml; /* MAC learning table. */
157 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
158 const struct ovsrec_bridge *cfg;
160 /* OpenFlow switch processing. */
161 struct ofproto *ofproto; /* OpenFlow switch. */
163 /* Kernel datapath information. */
164 struct dpif *dpif; /* Datapath. */
165 struct port_array ifaces; /* Indexed by kernel datapath port number. */
169 size_t n_ports, allocated_ports;
170 struct shash iface_by_name; /* "struct iface"s indexed by name. */
171 struct shash port_by_name; /* "struct port"s indexed by name. */
174 bool has_bonded_ports;
179 /* Port mirroring. */
180 struct mirror *mirrors[MAX_MIRRORS];
183 /* List of all bridges. */
184 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
186 /* OVSDB IDL used to obtain configuration. */
187 static struct ovsdb_idl *idl;
189 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
190 static void bridge_destroy(struct bridge *);
191 static struct bridge *bridge_lookup(const char *name);
192 static unixctl_cb_func bridge_unixctl_dump_flows;
193 static int bridge_run_one(struct bridge *);
194 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
195 const struct bridge *br,
196 struct ovsrec_controller ***controllersp);
197 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
199 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
201 const struct sockaddr_in *managers,
203 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
204 static void bridge_fetch_dp_ifaces(struct bridge *);
205 static void bridge_flush(struct bridge *);
206 static void bridge_pick_local_hw_addr(struct bridge *,
207 uint8_t ea[ETH_ADDR_LEN],
208 struct iface **hw_addr_iface);
209 static uint64_t bridge_pick_datapath_id(struct bridge *,
210 const uint8_t bridge_ea[ETH_ADDR_LEN],
211 struct iface *hw_addr_iface);
212 static struct iface *bridge_get_local_iface(struct bridge *);
213 static uint64_t dpid_from_hash(const void *, size_t nbytes);
215 static unixctl_cb_func bridge_unixctl_fdb_show;
217 static void bond_init(void);
218 static void bond_run(struct bridge *);
219 static void bond_wait(struct bridge *);
220 static void bond_rebalance_port(struct port *);
221 static void bond_send_learning_packets(struct port *);
222 static void bond_enable_slave(struct iface *iface, bool enable);
224 static struct port *port_create(struct bridge *, const char *name);
225 static void port_reconfigure(struct port *, const struct ovsrec_port *);
226 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
227 static void port_destroy(struct port *);
228 static struct port *port_lookup(const struct bridge *, const char *name);
229 static struct iface *port_lookup_iface(const struct port *, const char *name);
230 static struct port *port_from_dp_ifidx(const struct bridge *,
232 static void port_update_bond_compat(struct port *);
233 static void port_update_vlan_compat(struct port *);
234 static void port_update_bonding(struct port *);
236 static struct mirror *mirror_create(struct bridge *, const char *name);
237 static void mirror_destroy(struct mirror *);
238 static void mirror_reconfigure(struct bridge *);
239 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
240 static bool vlan_is_mirrored(const struct mirror *, int vlan);
242 static struct iface *iface_create(struct port *port,
243 const struct ovsrec_interface *if_cfg);
244 static void iface_destroy(struct iface *);
245 static struct iface *iface_lookup(const struct bridge *, const char *name);
246 static struct iface *iface_from_dp_ifidx(const struct bridge *,
248 static bool iface_is_internal(const struct bridge *, const char *name);
249 static void iface_set_mac(struct iface *);
250 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
252 /* Hooks into ofproto processing. */
253 static struct ofhooks bridge_ofhooks;
255 /* Public functions. */
257 /* Initializes the bridge module, configuring it to obtain its configuration
258 * from an OVSDB server accessed over 'remote', which should be a string in a
259 * form acceptable to ovsdb_idl_create(). */
261 bridge_init(const char *remote)
263 /* Create connection to database. */
264 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
266 /* Register unixctl commands. */
267 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
268 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
273 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
274 * but for which the ovs-vswitchd configuration 'cfg' is required. */
276 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
278 static bool already_configured_once;
279 struct svec bridge_names;
280 struct svec dpif_names, dpif_types;
283 /* Only do this once per ovs-vswitchd run. */
284 if (already_configured_once) {
287 already_configured_once = true;
289 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
290 svec_init(&bridge_names);
291 for (i = 0; i < cfg->n_bridges; i++) {
292 svec_add(&bridge_names, cfg->bridges[i]->name);
294 svec_sort(&bridge_names);
296 /* Iterate over all system dpifs and delete any of them that do not appear
298 svec_init(&dpif_names);
299 svec_init(&dpif_types);
300 dp_enumerate_types(&dpif_types);
301 for (i = 0; i < dpif_types.n; i++) {
306 dp_enumerate_names(dpif_types.names[i], &dpif_names);
308 /* For each dpif... */
309 for (j = 0; j < dpif_names.n; j++) {
310 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
312 struct svec all_names;
315 /* ...check whether any of its names is in 'bridge_names'. */
316 svec_init(&all_names);
317 dpif_get_all_names(dpif, &all_names);
318 for (k = 0; k < all_names.n; k++) {
319 if (svec_contains(&bridge_names, all_names.names[k])) {
324 /* No. Delete the dpif. */
328 svec_destroy(&all_names);
333 svec_destroy(&bridge_names);
334 svec_destroy(&dpif_names);
335 svec_destroy(&dpif_types);
340 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
342 /* XXX SSL should be configurable on a per-bridge basis. */
344 stream_ssl_set_private_key_file(ssl->private_key);
345 stream_ssl_set_certificate_file(ssl->certificate);
346 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
351 /* Attempt to create the network device 'iface_name' through the netdev
354 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
357 struct shash options;
361 shash_init(&options);
362 for (i = 0; i < iface_cfg->n_options; i++) {
363 shash_add(&options, iface_cfg->key_options[i],
364 xstrdup(iface_cfg->value_options[i]));
368 struct netdev_options netdev_options;
370 memset(&netdev_options, 0, sizeof netdev_options);
371 netdev_options.name = iface_cfg->name;
372 if (!strcmp(iface_cfg->type, "internal")) {
373 /* An "internal" config type maps to a netdev "system" type. */
374 netdev_options.type = "system";
376 netdev_options.type = iface_cfg->type;
378 netdev_options.args = &options;
379 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
381 error = netdev_open(&netdev_options, &iface->netdev);
384 netdev_get_carrier(iface->netdev, &iface->enabled);
386 } else if (iface->netdev) {
387 const char *netdev_type = netdev_get_type(iface->netdev);
388 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
389 ? iface_cfg->type : NULL;
391 /* An "internal" config type maps to a netdev "system" type. */
392 if (iface_type && !strcmp(iface_type, "internal")) {
393 iface_type = "system";
396 if (!iface_type || !strcmp(netdev_type, iface_type)) {
397 error = netdev_reconfigure(iface->netdev, &options);
399 VLOG_WARN("%s: attempting change device type from %s to %s",
400 iface_cfg->name, netdev_type, iface_type);
404 shash_destroy_free_data(&options);
410 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
412 return set_up_iface(iface_cfg, iface, false);
416 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
417 void *aux OVS_UNUSED)
419 if (!iface->netdev) {
420 int error = set_up_iface(iface->cfg, iface, true);
422 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
432 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
433 void *aux OVS_UNUSED)
435 if (iface->dp_ifidx >= 0) {
436 VLOG_DBG("%s has interface %s on port %d",
438 iface->name, iface->dp_ifidx);
441 VLOG_ERR("%s interface not in %s, dropping",
442 iface->name, dpif_name(br->dpif));
448 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
449 void *aux OVS_UNUSED)
451 /* Set policing attributes. */
452 netdev_set_policing(iface->netdev,
453 iface->cfg->ingress_policing_rate,
454 iface->cfg->ingress_policing_burst);
456 /* Set MAC address of internal interfaces other than the local
458 if (iface->dp_ifidx != ODPP_LOCAL
459 && iface_is_internal(br, iface->name)) {
460 iface_set_mac(iface);
466 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
467 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
468 * deletes from 'br' any ports that no longer have any interfaces. */
470 iterate_and_prune_ifaces(struct bridge *br,
471 bool (*cb)(struct bridge *, struct iface *,
477 for (i = 0; i < br->n_ports; ) {
478 struct port *port = br->ports[i];
479 for (j = 0; j < port->n_ifaces; ) {
480 struct iface *iface = port->ifaces[j];
481 if (cb(br, iface, aux)) {
484 iface_destroy(iface);
488 if (port->n_ifaces) {
491 VLOG_ERR("%s port has no interfaces, dropping", port->name);
497 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
498 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
499 * responsible for freeing '*managersp' (with free()).
501 * You may be asking yourself "why does ovs-vswitchd care?", because
502 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
503 * should not be and in fact is not directly involved in that. But
504 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
505 * it has to tell in-band control where the managers are to enable that.
508 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
509 struct sockaddr_in **managersp, size_t *n_managersp)
511 struct sockaddr_in *managers = NULL;
512 size_t n_managers = 0;
514 if (ovs_cfg->n_managers > 0) {
517 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
518 for (i = 0; i < ovs_cfg->n_managers; i++) {
519 const char *name = ovs_cfg->managers[i];
520 struct sockaddr_in *sin = &managers[i];
522 if ((!strncmp(name, "tcp:", 4)
523 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
524 (!strncmp(name, "ssl:", 4)
525 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
531 *managersp = managers;
532 *n_managersp = n_managers;
536 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
538 struct shash old_br, new_br;
539 struct shash_node *node;
540 struct bridge *br, *next;
541 struct sockaddr_in *managers;
544 int sflow_bridge_number;
546 COVERAGE_INC(bridge_reconfigure);
548 collect_managers(ovs_cfg, &managers, &n_managers);
550 /* Collect old and new bridges. */
553 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
554 shash_add(&old_br, br->name, br);
556 for (i = 0; i < ovs_cfg->n_bridges; i++) {
557 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
558 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
559 VLOG_WARN("more than one bridge named %s", br_cfg->name);
563 /* Get rid of deleted bridges and add new bridges. */
564 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
565 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
572 SHASH_FOR_EACH (node, &new_br) {
573 const char *br_name = node->name;
574 const struct ovsrec_bridge *br_cfg = node->data;
575 br = shash_find_data(&old_br, br_name);
577 /* If the bridge datapath type has changed, we need to tear it
578 * down and recreate. */
579 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
581 bridge_create(br_cfg);
584 bridge_create(br_cfg);
587 shash_destroy(&old_br);
588 shash_destroy(&new_br);
592 bridge_configure_ssl(ovs_cfg->ssl);
595 /* Reconfigure all bridges. */
596 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
597 bridge_reconfigure_one(ovs_cfg, br);
600 /* Add and delete ports on all datapaths.
602 * The kernel will reject any attempt to add a given port to a datapath if
603 * that port already belongs to a different datapath, so we must do all
604 * port deletions before any port additions. */
605 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
606 struct odp_port *dpif_ports;
608 struct shash want_ifaces;
610 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
611 bridge_get_all_ifaces(br, &want_ifaces);
612 for (i = 0; i < n_dpif_ports; i++) {
613 const struct odp_port *p = &dpif_ports[i];
614 if (!shash_find(&want_ifaces, p->devname)
615 && strcmp(p->devname, br->name)) {
616 int retval = dpif_port_del(br->dpif, p->port);
618 VLOG_ERR("failed to remove %s interface from %s: %s",
619 p->devname, dpif_name(br->dpif),
624 shash_destroy(&want_ifaces);
627 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
628 struct odp_port *dpif_ports;
630 struct shash cur_ifaces, want_ifaces;
631 struct shash_node *node;
633 /* Get the set of interfaces currently in this datapath. */
634 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
635 shash_init(&cur_ifaces);
636 for (i = 0; i < n_dpif_ports; i++) {
637 const char *name = dpif_ports[i].devname;
638 if (!shash_find(&cur_ifaces, name)) {
639 shash_add(&cur_ifaces, name, NULL);
644 /* Get the set of interfaces we want on this datapath. */
645 bridge_get_all_ifaces(br, &want_ifaces);
647 SHASH_FOR_EACH (node, &want_ifaces) {
648 const char *if_name = node->name;
649 struct iface *iface = node->data;
651 if (shash_find(&cur_ifaces, if_name)) {
652 /* Already exists, just reconfigure it. */
654 reconfigure_iface(iface->cfg, iface);
657 /* Need to add to datapath. */
661 /* Add to datapath. */
662 internal = iface_is_internal(br, if_name);
663 error = dpif_port_add(br->dpif, if_name,
664 internal ? ODP_PORT_INTERNAL : 0, NULL);
665 if (error == EFBIG) {
666 VLOG_ERR("ran out of valid port numbers on %s",
667 dpif_name(br->dpif));
670 VLOG_ERR("failed to add %s interface to %s: %s",
671 if_name, dpif_name(br->dpif), strerror(error));
675 shash_destroy(&cur_ifaces);
676 shash_destroy(&want_ifaces);
678 sflow_bridge_number = 0;
679 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
682 struct iface *local_iface;
683 struct iface *hw_addr_iface;
686 bridge_fetch_dp_ifaces(br);
688 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
689 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
691 /* Pick local port hardware address, datapath ID. */
692 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
693 local_iface = bridge_get_local_iface(br);
695 int error = netdev_set_etheraddr(local_iface->netdev, ea);
697 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
698 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
699 "Ethernet address: %s",
700 br->name, strerror(error));
704 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
705 ofproto_set_datapath_id(br->ofproto, dpid);
707 dpid_string = xasprintf("%016"PRIx64, dpid);
708 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
711 /* Set NetFlow configuration on this bridge. */
712 if (br->cfg->netflow) {
713 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
714 struct netflow_options opts;
716 memset(&opts, 0, sizeof opts);
718 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
719 if (nf_cfg->engine_type) {
720 opts.engine_type = *nf_cfg->engine_type;
722 if (nf_cfg->engine_id) {
723 opts.engine_id = *nf_cfg->engine_id;
726 opts.active_timeout = nf_cfg->active_timeout;
727 if (!opts.active_timeout) {
728 opts.active_timeout = -1;
729 } else if (opts.active_timeout < 0) {
730 VLOG_WARN("bridge %s: active timeout interval set to negative "
731 "value, using default instead (%d seconds)", br->name,
732 NF_ACTIVE_TIMEOUT_DEFAULT);
733 opts.active_timeout = -1;
736 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
737 if (opts.add_id_to_iface) {
738 if (opts.engine_id > 0x7f) {
739 VLOG_WARN("bridge %s: netflow port mangling may conflict "
740 "with another vswitch, choose an engine id less "
741 "than 128", br->name);
743 if (br->n_ports > 508) {
744 VLOG_WARN("bridge %s: netflow port mangling will conflict "
745 "with another port when more than 508 ports are "
750 opts.collectors.n = nf_cfg->n_targets;
751 opts.collectors.names = nf_cfg->targets;
752 if (ofproto_set_netflow(br->ofproto, &opts)) {
753 VLOG_ERR("bridge %s: problem setting netflow collectors",
757 ofproto_set_netflow(br->ofproto, NULL);
760 /* Set sFlow configuration on this bridge. */
761 if (br->cfg->sflow) {
762 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
763 struct ovsrec_controller **controllers;
764 struct ofproto_sflow_options oso;
765 size_t n_controllers;
768 memset(&oso, 0, sizeof oso);
770 oso.targets.n = sflow_cfg->n_targets;
771 oso.targets.names = sflow_cfg->targets;
773 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
774 if (sflow_cfg->sampling) {
775 oso.sampling_rate = *sflow_cfg->sampling;
778 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
779 if (sflow_cfg->polling) {
780 oso.polling_interval = *sflow_cfg->polling;
783 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
784 if (sflow_cfg->header) {
785 oso.header_len = *sflow_cfg->header;
788 oso.sub_id = sflow_bridge_number++;
789 oso.agent_device = sflow_cfg->agent;
791 oso.control_ip = NULL;
792 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
793 for (i = 0; i < n_controllers; i++) {
794 if (controllers[i]->local_ip) {
795 oso.control_ip = controllers[i]->local_ip;
799 ofproto_set_sflow(br->ofproto, &oso);
801 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
803 ofproto_set_sflow(br->ofproto, NULL);
806 /* Update the controller and related settings. It would be more
807 * straightforward to call this from bridge_reconfigure_one(), but we
808 * can't do it there for two reasons. First, and most importantly, at
809 * that point we don't know the dp_ifidx of any interfaces that have
810 * been added to the bridge (because we haven't actually added them to
811 * the datapath). Second, at that point we haven't set the datapath ID
812 * yet; when a controller is configured, resetting the datapath ID will
813 * immediately disconnect from the controller, so it's better to set
814 * the datapath ID before the controller. */
815 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
817 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
818 for (i = 0; i < br->n_ports; i++) {
819 struct port *port = br->ports[i];
822 port_update_vlan_compat(port);
823 port_update_bonding(port);
825 for (j = 0; j < port->n_ifaces; j++) {
826 iface_update_qos(port->ifaces[j], port->cfg->qos);
830 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
831 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
838 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
842 for (i = 0; i < n; i++) {
843 if (!strcmp(keys[i], key)) {
851 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
853 return get_ovsrec_key_value(key,
854 br_cfg->key_other_config,
855 br_cfg->value_other_config,
856 br_cfg->n_other_config);
860 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
861 struct iface **hw_addr_iface)
867 *hw_addr_iface = NULL;
869 /* Did the user request a particular MAC? */
870 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
871 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
872 if (eth_addr_is_multicast(ea)) {
873 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
874 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
875 } else if (eth_addr_is_zero(ea)) {
876 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
882 /* Otherwise choose the minimum non-local MAC address among all of the
884 memset(ea, 0xff, sizeof ea);
885 for (i = 0; i < br->n_ports; i++) {
886 struct port *port = br->ports[i];
887 uint8_t iface_ea[ETH_ADDR_LEN];
890 /* Mirror output ports don't participate. */
891 if (port->is_mirror_output_port) {
895 /* Choose the MAC address to represent the port. */
896 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
897 /* Find the interface with this Ethernet address (if any) so that
898 * we can provide the correct devname to the caller. */
900 for (j = 0; j < port->n_ifaces; j++) {
901 struct iface *candidate = port->ifaces[j];
902 uint8_t candidate_ea[ETH_ADDR_LEN];
903 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
904 && eth_addr_equals(iface_ea, candidate_ea)) {
909 /* Choose the interface whose MAC address will represent the port.
910 * The Linux kernel bonding code always chooses the MAC address of
911 * the first slave added to a bond, and the Fedora networking
912 * scripts always add slaves to a bond in alphabetical order, so
913 * for compatibility we choose the interface with the name that is
914 * first in alphabetical order. */
915 iface = port->ifaces[0];
916 for (j = 1; j < port->n_ifaces; j++) {
917 struct iface *candidate = port->ifaces[j];
918 if (strcmp(candidate->name, iface->name) < 0) {
923 /* The local port doesn't count (since we're trying to choose its
924 * MAC address anyway). */
925 if (iface->dp_ifidx == ODPP_LOCAL) {
930 error = netdev_get_etheraddr(iface->netdev, iface_ea);
932 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
933 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
934 iface->name, strerror(error));
939 /* Compare against our current choice. */
940 if (!eth_addr_is_multicast(iface_ea) &&
941 !eth_addr_is_local(iface_ea) &&
942 !eth_addr_is_reserved(iface_ea) &&
943 !eth_addr_is_zero(iface_ea) &&
944 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
946 memcpy(ea, iface_ea, ETH_ADDR_LEN);
947 *hw_addr_iface = iface;
950 if (eth_addr_is_multicast(ea)) {
951 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
952 *hw_addr_iface = NULL;
953 VLOG_WARN("bridge %s: using default bridge Ethernet "
954 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
956 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
957 br->name, ETH_ADDR_ARGS(ea));
961 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
962 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
963 * an interface on 'br', then that interface must be passed in as
964 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
965 * 'hw_addr_iface' must be passed in as a null pointer. */
967 bridge_pick_datapath_id(struct bridge *br,
968 const uint8_t bridge_ea[ETH_ADDR_LEN],
969 struct iface *hw_addr_iface)
972 * The procedure for choosing a bridge MAC address will, in the most
973 * ordinary case, also choose a unique MAC that we can use as a datapath
974 * ID. In some special cases, though, multiple bridges will end up with
975 * the same MAC address. This is OK for the bridges, but it will confuse
976 * the OpenFlow controller, because each datapath needs a unique datapath
979 * Datapath IDs must be unique. It is also very desirable that they be
980 * stable from one run to the next, so that policy set on a datapath
983 const char *datapath_id;
986 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
987 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
993 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
995 * A bridge whose MAC address is taken from a VLAN network device
996 * (that is, a network device created with vconfig(8) or similar
997 * tool) will have the same MAC address as a bridge on the VLAN
998 * device's physical network device.
1000 * Handle this case by hashing the physical network device MAC
1001 * along with the VLAN identifier.
1003 uint8_t buf[ETH_ADDR_LEN + 2];
1004 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1005 buf[ETH_ADDR_LEN] = vlan >> 8;
1006 buf[ETH_ADDR_LEN + 1] = vlan;
1007 return dpid_from_hash(buf, sizeof buf);
1010 * Assume that this bridge's MAC address is unique, since it
1011 * doesn't fit any of the cases we handle specially.
1016 * A purely internal bridge, that is, one that has no non-virtual
1017 * network devices on it at all, is more difficult because it has no
1018 * natural unique identifier at all.
1020 * When the host is a XenServer, we handle this case by hashing the
1021 * host's UUID with the name of the bridge. Names of bridges are
1022 * persistent across XenServer reboots, although they can be reused if
1023 * an internal network is destroyed and then a new one is later
1024 * created, so this is fairly effective.
1026 * When the host is not a XenServer, we punt by using a random MAC
1027 * address on each run.
1029 const char *host_uuid = xenserver_get_host_uuid();
1031 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1032 dpid = dpid_from_hash(combined, strlen(combined));
1038 return eth_addr_to_uint64(bridge_ea);
1042 dpid_from_hash(const void *data, size_t n)
1044 uint8_t hash[SHA1_DIGEST_SIZE];
1046 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1047 sha1_bytes(data, n, hash);
1048 eth_addr_mark_random(hash);
1049 return eth_addr_to_uint64(hash);
1055 bool datapath_destroyed;
1058 /* Let each bridge do the work that it needs to do. */
1059 datapath_destroyed = false;
1060 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1061 int error = bridge_run_one(br);
1063 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1064 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1065 "forcing reconfiguration", br->name);
1066 datapath_destroyed = true;
1070 /* (Re)configure if necessary. */
1071 if (ovsdb_idl_run(idl) || datapath_destroyed) {
1072 const struct ovsrec_open_vswitch *cfg = ovsrec_open_vswitch_first(idl);
1074 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1076 bridge_configure_once(cfg);
1077 bridge_reconfigure(cfg);
1079 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1080 ovsdb_idl_txn_commit(txn);
1081 ovsdb_idl_txn_destroy(txn); /* XXX */
1083 /* We still need to reconfigure to avoid dangling pointers to
1084 * now-destroyed ovsrec structures inside bridge data. */
1085 static const struct ovsrec_open_vswitch null_cfg;
1087 bridge_reconfigure(&null_cfg);
1097 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1098 ofproto_wait(br->ofproto);
1099 if (ofproto_has_controller(br->ofproto)) {
1103 mac_learning_wait(br->ml);
1106 ovsdb_idl_wait(idl);
1109 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1110 * configuration changes. */
1112 bridge_flush(struct bridge *br)
1114 COVERAGE_INC(bridge_flush);
1116 mac_learning_flush(br->ml);
1119 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1120 * such interface. */
1121 static struct iface *
1122 bridge_get_local_iface(struct bridge *br)
1126 for (i = 0; i < br->n_ports; i++) {
1127 struct port *port = br->ports[i];
1128 for (j = 0; j < port->n_ifaces; j++) {
1129 struct iface *iface = port->ifaces[j];
1130 if (iface->dp_ifidx == ODPP_LOCAL) {
1139 /* Bridge unixctl user interface functions. */
1141 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1142 const char *args, void *aux OVS_UNUSED)
1144 struct ds ds = DS_EMPTY_INITIALIZER;
1145 const struct bridge *br;
1146 const struct mac_entry *e;
1148 br = bridge_lookup(args);
1150 unixctl_command_reply(conn, 501, "no such bridge");
1154 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1155 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1156 if (e->port < 0 || e->port >= br->n_ports) {
1159 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1160 br->ports[e->port]->ifaces[0]->dp_ifidx,
1161 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1163 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1167 /* Bridge reconfiguration functions. */
1168 static struct bridge *
1169 bridge_create(const struct ovsrec_bridge *br_cfg)
1174 assert(!bridge_lookup(br_cfg->name));
1175 br = xzalloc(sizeof *br);
1177 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1183 dpif_flow_flush(br->dpif);
1185 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1188 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1190 dpif_delete(br->dpif);
1191 dpif_close(br->dpif);
1196 br->name = xstrdup(br_cfg->name);
1198 br->ml = mac_learning_create();
1199 eth_addr_nicira_random(br->default_ea);
1201 port_array_init(&br->ifaces);
1203 shash_init(&br->port_by_name);
1204 shash_init(&br->iface_by_name);
1208 list_push_back(&all_bridges, &br->node);
1210 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1216 bridge_destroy(struct bridge *br)
1221 while (br->n_ports > 0) {
1222 port_destroy(br->ports[br->n_ports - 1]);
1224 list_remove(&br->node);
1225 error = dpif_delete(br->dpif);
1226 if (error && error != ENOENT) {
1227 VLOG_ERR("failed to delete %s: %s",
1228 dpif_name(br->dpif), strerror(error));
1230 dpif_close(br->dpif);
1231 ofproto_destroy(br->ofproto);
1232 mac_learning_destroy(br->ml);
1233 port_array_destroy(&br->ifaces);
1234 shash_destroy(&br->port_by_name);
1235 shash_destroy(&br->iface_by_name);
1242 static struct bridge *
1243 bridge_lookup(const char *name)
1247 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1248 if (!strcmp(br->name, name)) {
1255 /* Handle requests for a listing of all flows known by the OpenFlow
1256 * stack, including those normally hidden. */
1258 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1259 const char *args, void *aux OVS_UNUSED)
1264 br = bridge_lookup(args);
1266 unixctl_command_reply(conn, 501, "Unknown bridge");
1271 ofproto_get_all_flows(br->ofproto, &results);
1273 unixctl_command_reply(conn, 200, ds_cstr(&results));
1274 ds_destroy(&results);
1278 bridge_run_one(struct bridge *br)
1282 error = ofproto_run1(br->ofproto);
1287 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1290 error = ofproto_run2(br->ofproto, br->flush);
1297 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1298 const struct bridge *br,
1299 struct ovsrec_controller ***controllersp)
1301 struct ovsrec_controller **controllers;
1302 size_t n_controllers;
1304 if (br->cfg->n_controller) {
1305 controllers = br->cfg->controller;
1306 n_controllers = br->cfg->n_controller;
1308 controllers = ovs_cfg->controller;
1309 n_controllers = ovs_cfg->n_controller;
1312 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1318 *controllersp = controllers;
1320 return n_controllers;
1324 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1327 struct shash old_ports, new_ports;
1328 struct svec listeners, old_listeners;
1329 struct svec snoops, old_snoops;
1330 struct shash_node *node;
1333 /* Collect old ports. */
1334 shash_init(&old_ports);
1335 for (i = 0; i < br->n_ports; i++) {
1336 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1339 /* Collect new ports. */
1340 shash_init(&new_ports);
1341 for (i = 0; i < br->cfg->n_ports; i++) {
1342 const char *name = br->cfg->ports[i]->name;
1343 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1344 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1349 /* If we have a controller, then we need a local port. Complain if the
1350 * user didn't specify one.
1352 * XXX perhaps we should synthesize a port ourselves in this case. */
1353 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1354 char local_name[IF_NAMESIZE];
1357 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1358 local_name, sizeof local_name);
1359 if (!error && !shash_find(&new_ports, local_name)) {
1360 VLOG_WARN("bridge %s: controller specified but no local port "
1361 "(port named %s) defined",
1362 br->name, local_name);
1366 /* Get rid of deleted ports.
1367 * Get rid of deleted interfaces on ports that still exist. */
1368 SHASH_FOR_EACH (node, &old_ports) {
1369 struct port *port = node->data;
1370 const struct ovsrec_port *port_cfg;
1372 port_cfg = shash_find_data(&new_ports, node->name);
1376 port_del_ifaces(port, port_cfg);
1380 /* Create new ports.
1381 * Add new interfaces to existing ports.
1382 * Reconfigure existing ports. */
1383 SHASH_FOR_EACH (node, &new_ports) {
1384 struct port *port = shash_find_data(&old_ports, node->name);
1386 port = port_create(br, node->name);
1389 port_reconfigure(port, node->data);
1390 if (!port->n_ifaces) {
1391 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1392 br->name, port->name);
1396 shash_destroy(&old_ports);
1397 shash_destroy(&new_ports);
1399 /* Delete all flows if we're switching from connected to standalone or vice
1400 * versa. (XXX Should we delete all flows if we are switching from one
1401 * controller to another?) */
1403 /* Configure OpenFlow management listener. */
1404 svec_init(&listeners);
1405 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1406 ovs_rundir, br->name));
1407 svec_init(&old_listeners);
1408 ofproto_get_listeners(br->ofproto, &old_listeners);
1409 if (!svec_equal(&listeners, &old_listeners)) {
1410 ofproto_set_listeners(br->ofproto, &listeners);
1412 svec_destroy(&listeners);
1413 svec_destroy(&old_listeners);
1415 /* Configure OpenFlow controller connection snooping. */
1417 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1418 ovs_rundir, br->name));
1419 svec_init(&old_snoops);
1420 ofproto_get_snoops(br->ofproto, &old_snoops);
1421 if (!svec_equal(&snoops, &old_snoops)) {
1422 ofproto_set_snoops(br->ofproto, &snoops);
1424 svec_destroy(&snoops);
1425 svec_destroy(&old_snoops);
1427 mirror_reconfigure(br);
1431 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1433 const struct sockaddr_in *managers,
1436 struct ovsrec_controller **controllers;
1437 size_t n_controllers;
1439 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1441 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1442 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1443 ofproto_flush_flows(br->ofproto);
1446 if (!n_controllers) {
1447 union ofp_action action;
1450 /* Clear out controllers. */
1451 ofproto_set_controllers(br->ofproto, NULL, 0);
1453 /* Set up a flow that matches every packet and directs them to
1454 * OFPP_NORMAL (which goes to us). */
1455 memset(&action, 0, sizeof action);
1456 action.type = htons(OFPAT_OUTPUT);
1457 action.output.len = htons(sizeof action);
1458 action.output.port = htons(OFPP_NORMAL);
1459 memset(&flow, 0, sizeof flow);
1460 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1462 struct ofproto_controller *ocs;
1465 ocs = xmalloc(n_controllers * sizeof *ocs);
1466 for (i = 0; i < n_controllers; i++) {
1467 struct ovsrec_controller *c = controllers[i];
1468 struct ofproto_controller *oc = &ocs[i];
1470 if (strcmp(c->target, "discover")) {
1471 struct iface *local_iface;
1474 local_iface = bridge_get_local_iface(br);
1475 if (local_iface && c->local_ip
1476 && inet_aton(c->local_ip, &ip)) {
1477 struct netdev *netdev = local_iface->netdev;
1478 struct in_addr mask, gateway;
1480 if (!c->local_netmask
1481 || !inet_aton(c->local_netmask, &mask)) {
1484 if (!c->local_gateway
1485 || !inet_aton(c->local_gateway, &gateway)) {
1489 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1491 mask.s_addr = guess_netmask(ip.s_addr);
1493 if (!netdev_set_in4(netdev, ip, mask)) {
1494 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1496 br->name, IP_ARGS(&ip.s_addr),
1497 IP_ARGS(&mask.s_addr));
1500 if (gateway.s_addr) {
1501 if (!netdev_add_router(netdev, gateway)) {
1502 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1503 br->name, IP_ARGS(&gateway.s_addr));
1509 oc->target = c->target;
1510 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1511 oc->probe_interval = (c->inactivity_probe
1512 ? *c->inactivity_probe / 1000 : 5);
1513 oc->fail = (!c->fail_mode
1514 || !strcmp(c->fail_mode, "standalone")
1515 || !strcmp(c->fail_mode, "open")
1516 ? OFPROTO_FAIL_STANDALONE
1517 : OFPROTO_FAIL_SECURE);
1518 oc->band = (!c->connection_mode
1519 || !strcmp(c->connection_mode, "in-band")
1521 : OFPROTO_OUT_OF_BAND);
1522 oc->accept_re = c->discover_accept_regex;
1523 oc->update_resolv_conf = c->discover_update_resolv_conf;
1524 oc->rate_limit = (c->controller_rate_limit
1525 ? *c->controller_rate_limit : 0);
1526 oc->burst_limit = (c->controller_burst_limit
1527 ? *c->controller_burst_limit : 0);
1529 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1535 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1540 for (i = 0; i < br->n_ports; i++) {
1541 struct port *port = br->ports[i];
1542 for (j = 0; j < port->n_ifaces; j++) {
1543 struct iface *iface = port->ifaces[j];
1544 shash_add_once(ifaces, iface->name, iface);
1546 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1547 shash_add_once(ifaces, port->name, NULL);
1552 /* For robustness, in case the administrator moves around datapath ports behind
1553 * our back, we re-check all the datapath port numbers here.
1555 * This function will set the 'dp_ifidx' members of interfaces that have
1556 * disappeared to -1, so only call this function from a context where those
1557 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1558 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1559 * datapath, which doesn't support UINT16_MAX+1 ports. */
1561 bridge_fetch_dp_ifaces(struct bridge *br)
1563 struct odp_port *dpif_ports;
1564 size_t n_dpif_ports;
1567 /* Reset all interface numbers. */
1568 for (i = 0; i < br->n_ports; i++) {
1569 struct port *port = br->ports[i];
1570 for (j = 0; j < port->n_ifaces; j++) {
1571 struct iface *iface = port->ifaces[j];
1572 iface->dp_ifidx = -1;
1575 port_array_clear(&br->ifaces);
1577 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1578 for (i = 0; i < n_dpif_ports; i++) {
1579 struct odp_port *p = &dpif_ports[i];
1580 struct iface *iface = iface_lookup(br, p->devname);
1582 if (iface->dp_ifidx >= 0) {
1583 VLOG_WARN("%s reported interface %s twice",
1584 dpif_name(br->dpif), p->devname);
1585 } else if (iface_from_dp_ifidx(br, p->port)) {
1586 VLOG_WARN("%s reported interface %"PRIu16" twice",
1587 dpif_name(br->dpif), p->port);
1589 port_array_set(&br->ifaces, p->port, iface);
1590 iface->dp_ifidx = p->port;
1594 int64_t ofport = (iface->dp_ifidx >= 0
1595 ? odp_port_to_ofp_port(iface->dp_ifidx)
1597 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1604 /* Bridge packet processing functions. */
1607 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1609 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1612 static struct bond_entry *
1613 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1615 return &port->bond_hash[bond_hash(mac)];
1619 bond_choose_iface(const struct port *port)
1621 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1622 size_t i, best_down_slave = -1;
1623 long long next_delay_expiration = LLONG_MAX;
1625 for (i = 0; i < port->n_ifaces; i++) {
1626 struct iface *iface = port->ifaces[i];
1628 if (iface->enabled) {
1630 } else if (iface->delay_expires < next_delay_expiration) {
1631 best_down_slave = i;
1632 next_delay_expiration = iface->delay_expires;
1636 if (best_down_slave != -1) {
1637 struct iface *iface = port->ifaces[best_down_slave];
1639 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1640 "since no other interface is up", iface->name,
1641 iface->delay_expires - time_msec());
1642 bond_enable_slave(iface, true);
1645 return best_down_slave;
1649 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1650 uint16_t *dp_ifidx, tag_type *tags)
1652 struct iface *iface;
1654 assert(port->n_ifaces);
1655 if (port->n_ifaces == 1) {
1656 iface = port->ifaces[0];
1658 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1659 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1660 || !port->ifaces[e->iface_idx]->enabled) {
1661 /* XXX select interface properly. The current interface selection
1662 * is only good for testing the rebalancing code. */
1663 e->iface_idx = bond_choose_iface(port);
1664 if (e->iface_idx < 0) {
1665 *tags |= port->no_ifaces_tag;
1668 e->iface_tag = tag_create_random();
1669 ((struct port *) port)->bond_compat_is_stale = true;
1671 *tags |= e->iface_tag;
1672 iface = port->ifaces[e->iface_idx];
1674 *dp_ifidx = iface->dp_ifidx;
1675 *tags |= iface->tag; /* Currently only used for bonding. */
1680 bond_link_status_update(struct iface *iface, bool carrier)
1682 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1683 struct port *port = iface->port;
1685 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1686 /* Nothing to do. */
1689 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1690 iface->name, carrier ? "detected" : "dropped");
1691 if (carrier == iface->enabled) {
1692 iface->delay_expires = LLONG_MAX;
1693 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1694 iface->name, carrier ? "disabled" : "enabled");
1695 } else if (carrier && port->active_iface < 0) {
1696 bond_enable_slave(iface, true);
1697 if (port->updelay) {
1698 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1699 "other interface is up", iface->name, port->updelay);
1702 int delay = carrier ? port->updelay : port->downdelay;
1703 iface->delay_expires = time_msec() + delay;
1706 "interface %s: will be %s if it stays %s for %d ms",
1708 carrier ? "enabled" : "disabled",
1709 carrier ? "up" : "down",
1716 bond_choose_active_iface(struct port *port)
1718 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1720 port->active_iface = bond_choose_iface(port);
1721 port->active_iface_tag = tag_create_random();
1722 if (port->active_iface >= 0) {
1723 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1724 port->name, port->ifaces[port->active_iface]->name);
1726 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1732 bond_enable_slave(struct iface *iface, bool enable)
1734 struct port *port = iface->port;
1735 struct bridge *br = port->bridge;
1737 /* This acts as a recursion check. If the act of disabling a slave
1738 * causes a different slave to be enabled, the flag will allow us to
1739 * skip redundant work when we reenter this function. It must be
1740 * cleared on exit to keep things safe with multiple bonds. */
1741 static bool moving_active_iface = false;
1743 iface->delay_expires = LLONG_MAX;
1744 if (enable == iface->enabled) {
1748 iface->enabled = enable;
1749 if (!iface->enabled) {
1750 VLOG_WARN("interface %s: disabled", iface->name);
1751 ofproto_revalidate(br->ofproto, iface->tag);
1752 if (iface->port_ifidx == port->active_iface) {
1753 ofproto_revalidate(br->ofproto,
1754 port->active_iface_tag);
1756 /* Disabling a slave can lead to another slave being immediately
1757 * enabled if there will be no active slaves but one is waiting
1758 * on an updelay. In this case we do not need to run most of the
1759 * code for the newly enabled slave since there was no period
1760 * without an active slave and it is redundant with the disabling
1762 moving_active_iface = true;
1763 bond_choose_active_iface(port);
1765 bond_send_learning_packets(port);
1767 VLOG_WARN("interface %s: enabled", iface->name);
1768 if (port->active_iface < 0 && !moving_active_iface) {
1769 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1770 bond_choose_active_iface(port);
1771 bond_send_learning_packets(port);
1773 iface->tag = tag_create_random();
1776 moving_active_iface = false;
1777 port->bond_compat_is_stale = true;
1780 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1781 * bond interface. */
1783 bond_update_fake_iface_stats(struct port *port)
1785 struct netdev_stats bond_stats;
1786 struct netdev *bond_dev;
1789 memset(&bond_stats, 0, sizeof bond_stats);
1791 for (i = 0; i < port->n_ifaces; i++) {
1792 struct netdev_stats slave_stats;
1794 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1795 /* XXX: We swap the stats here because they are swapped back when
1796 * reported by the internal device. The reason for this is
1797 * internal devices normally represent packets going into the system
1798 * but when used as fake bond device they represent packets leaving
1799 * the system. We really should do this in the internal device
1800 * itself because changing it here reverses the counts from the
1801 * perspective of the switch. However, the internal device doesn't
1802 * know what type of device it represents so we have to do it here
1804 bond_stats.tx_packets += slave_stats.rx_packets;
1805 bond_stats.tx_bytes += slave_stats.rx_bytes;
1806 bond_stats.rx_packets += slave_stats.tx_packets;
1807 bond_stats.rx_bytes += slave_stats.tx_bytes;
1811 if (!netdev_open_default(port->name, &bond_dev)) {
1812 netdev_set_stats(bond_dev, &bond_stats);
1813 netdev_close(bond_dev);
1818 bond_run(struct bridge *br)
1822 for (i = 0; i < br->n_ports; i++) {
1823 struct port *port = br->ports[i];
1825 if (port->n_ifaces >= 2) {
1826 for (j = 0; j < port->n_ifaces; j++) {
1827 struct iface *iface = port->ifaces[j];
1828 if (time_msec() >= iface->delay_expires) {
1829 bond_enable_slave(iface, !iface->enabled);
1833 if (port->bond_fake_iface
1834 && time_msec() >= port->bond_next_fake_iface_update) {
1835 bond_update_fake_iface_stats(port);
1836 port->bond_next_fake_iface_update = time_msec() + 1000;
1840 if (port->bond_compat_is_stale) {
1841 port->bond_compat_is_stale = false;
1842 port_update_bond_compat(port);
1848 bond_wait(struct bridge *br)
1852 for (i = 0; i < br->n_ports; i++) {
1853 struct port *port = br->ports[i];
1854 if (port->n_ifaces < 2) {
1857 for (j = 0; j < port->n_ifaces; j++) {
1858 struct iface *iface = port->ifaces[j];
1859 if (iface->delay_expires != LLONG_MAX) {
1860 poll_timer_wait_until(iface->delay_expires);
1863 if (port->bond_fake_iface) {
1864 poll_timer_wait_until(port->bond_next_fake_iface_update);
1870 set_dst(struct dst *p, const flow_t *flow,
1871 const struct port *in_port, const struct port *out_port,
1874 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1875 : in_port->vlan >= 0 ? in_port->vlan
1876 : ntohs(flow->dl_vlan));
1877 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1881 swap_dst(struct dst *p, struct dst *q)
1883 struct dst tmp = *p;
1888 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1889 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1890 * that we push to the datapath. We could in fact fully sort the array by
1891 * vlan, but in most cases there are at most two different vlan tags so that's
1892 * possibly overkill.) */
1894 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1896 struct dst *first = dsts;
1897 struct dst *last = dsts + n_dsts;
1899 while (first != last) {
1901 * - All dsts < first have vlan == 'vlan'.
1902 * - All dsts >= last have vlan != 'vlan'.
1903 * - first < last. */
1904 while (first->vlan == vlan) {
1905 if (++first == last) {
1910 /* Same invariants, plus one additional:
1911 * - first->vlan != vlan.
1913 while (last[-1].vlan != vlan) {
1914 if (--last == first) {
1919 /* Same invariants, plus one additional:
1920 * - last[-1].vlan == vlan.*/
1921 swap_dst(first++, --last);
1926 mirror_mask_ffs(mirror_mask_t mask)
1928 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1933 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1934 const struct dst *test)
1937 for (i = 0; i < n_dsts; i++) {
1938 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1946 port_trunks_vlan(const struct port *port, uint16_t vlan)
1948 return (port->vlan < 0
1949 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
1953 port_includes_vlan(const struct port *port, uint16_t vlan)
1955 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1959 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1960 const struct port *in_port, const struct port *out_port,
1961 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1963 mirror_mask_t mirrors = in_port->src_mirrors;
1964 struct dst *dst = dsts;
1967 if (out_port == FLOOD_PORT) {
1968 /* XXX use ODP_FLOOD if no vlans or bonding. */
1969 /* XXX even better, define each VLAN as a datapath port group */
1970 for (i = 0; i < br->n_ports; i++) {
1971 struct port *port = br->ports[i];
1972 if (port != in_port && port_includes_vlan(port, vlan)
1973 && !port->is_mirror_output_port
1974 && set_dst(dst, flow, in_port, port, tags)) {
1975 mirrors |= port->dst_mirrors;
1979 *nf_output_iface = NF_OUT_FLOOD;
1980 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1981 *nf_output_iface = dst->dp_ifidx;
1982 mirrors |= out_port->dst_mirrors;
1987 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1988 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1990 if (set_dst(dst, flow, in_port, m->out_port, tags)
1991 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1995 for (i = 0; i < br->n_ports; i++) {
1996 struct port *port = br->ports[i];
1997 if (port_includes_vlan(port, m->out_vlan)
1998 && set_dst(dst, flow, in_port, port, tags))
2002 if (port->vlan < 0) {
2003 dst->vlan = m->out_vlan;
2005 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2009 /* Use the vlan tag on the original flow instead of
2010 * the one passed in the vlan parameter. This ensures
2011 * that we compare the vlan from before any implicit
2012 * tagging tags place. This is necessary because
2013 * dst->vlan is the final vlan, after removing implicit
2015 flow_vlan = ntohs(flow->dl_vlan);
2016 if (flow_vlan == 0) {
2017 flow_vlan = OFP_VLAN_NONE;
2019 if (port == in_port && dst->vlan == flow_vlan) {
2020 /* Don't send out input port on same VLAN. */
2028 mirrors &= mirrors - 1;
2031 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2035 static void OVS_UNUSED
2036 print_dsts(const struct dst *dsts, size_t n)
2038 for (; n--; dsts++) {
2039 printf(">p%"PRIu16, dsts->dp_ifidx);
2040 if (dsts->vlan != OFP_VLAN_NONE) {
2041 printf("v%"PRIu16, dsts->vlan);
2047 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2048 const struct port *in_port, const struct port *out_port,
2049 tag_type *tags, struct odp_actions *actions,
2050 uint16_t *nf_output_iface)
2052 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2054 const struct dst *p;
2057 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2060 cur_vlan = ntohs(flow->dl_vlan);
2061 for (p = dsts; p < &dsts[n_dsts]; p++) {
2062 union odp_action *a;
2063 if (p->vlan != cur_vlan) {
2064 if (p->vlan == OFP_VLAN_NONE) {
2065 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2067 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2068 a->vlan_vid.vlan_vid = htons(p->vlan);
2072 a = odp_actions_add(actions, ODPAT_OUTPUT);
2073 a->output.port = p->dp_ifidx;
2077 /* Returns the effective vlan of a packet, taking into account both the
2078 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2079 * the packet is untagged and -1 indicates it has an invalid header and
2080 * should be dropped. */
2081 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2082 struct port *in_port, bool have_packet)
2084 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2085 * belongs to VLAN 0, so we should treat both cases identically. (In the
2086 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2087 * presumably to allow a priority to be specified. In the latter case, the
2088 * packet does not have any 802.1Q header.) */
2089 int vlan = ntohs(flow->dl_vlan);
2090 if (vlan == OFP_VLAN_NONE) {
2093 if (in_port->vlan >= 0) {
2095 /* XXX support double tagging? */
2097 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2098 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2099 "packet received on port %s configured with "
2100 "implicit VLAN %"PRIu16,
2101 br->name, ntohs(flow->dl_vlan),
2102 in_port->name, in_port->vlan);
2106 vlan = in_port->vlan;
2108 if (!port_includes_vlan(in_port, vlan)) {
2110 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2111 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2112 "packet received on port %s not configured for "
2114 br->name, vlan, in_port->name, vlan);
2123 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2124 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2125 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2127 is_gratuitous_arp(const flow_t *flow)
2129 return (flow->dl_type == htons(ETH_TYPE_ARP)
2130 && eth_addr_is_broadcast(flow->dl_dst)
2131 && (flow->nw_proto == ARP_OP_REPLY
2132 || (flow->nw_proto == ARP_OP_REQUEST
2133 && flow->nw_src == flow->nw_dst)));
2137 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2138 struct port *in_port)
2140 enum grat_arp_lock_type lock_type;
2143 /* We don't want to learn from gratuitous ARP packets that are reflected
2144 * back over bond slaves so we lock the learning table. */
2145 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2146 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2147 GRAT_ARP_LOCK_CHECK;
2149 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2152 /* The log messages here could actually be useful in debugging,
2153 * so keep the rate limit relatively high. */
2154 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2156 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2157 "on port %s in VLAN %d",
2158 br->name, ETH_ADDR_ARGS(flow->dl_src),
2159 in_port->name, vlan);
2160 ofproto_revalidate(br->ofproto, rev_tag);
2164 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2165 * dropped. Returns true if they may be forwarded, false if they should be
2168 * If 'have_packet' is true, it indicates that the caller is processing a
2169 * received packet. If 'have_packet' is false, then the caller is just
2170 * revalidating an existing flow because configuration has changed. Either
2171 * way, 'have_packet' only affects logging (there is no point in logging errors
2172 * during revalidation).
2174 * Sets '*in_portp' to the input port. This will be a null pointer if
2175 * flow->in_port does not designate a known input port (in which case
2176 * is_admissible() returns false).
2178 * When returning true, sets '*vlanp' to the effective VLAN of the input
2179 * packet, as returned by flow_get_vlan().
2181 * May also add tags to '*tags', although the current implementation only does
2182 * so in one special case.
2185 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2186 tag_type *tags, int *vlanp, struct port **in_portp)
2188 struct iface *in_iface;
2189 struct port *in_port;
2192 /* Find the interface and port structure for the received packet. */
2193 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2195 /* No interface? Something fishy... */
2197 /* Odd. A few possible reasons here:
2199 * - We deleted an interface but there are still a few packets
2200 * queued up from it.
2202 * - Someone externally added an interface (e.g. with "ovs-dpctl
2203 * add-if") that we don't know about.
2205 * - Packet arrived on the local port but the local port is not
2206 * one of our bridge ports.
2208 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2210 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2211 "interface %"PRIu16, br->name, flow->in_port);
2217 *in_portp = in_port = in_iface->port;
2218 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2223 /* Drop frames for reserved multicast addresses. */
2224 if (eth_addr_is_reserved(flow->dl_dst)) {
2228 /* Drop frames on ports reserved for mirroring. */
2229 if (in_port->is_mirror_output_port) {
2231 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2232 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2233 "%s, which is reserved exclusively for mirroring",
2234 br->name, in_port->name);
2239 /* Packets received on bonds need special attention to avoid duplicates. */
2240 if (in_port->n_ifaces > 1) {
2242 bool is_grat_arp_locked;
2244 if (eth_addr_is_multicast(flow->dl_dst)) {
2245 *tags |= in_port->active_iface_tag;
2246 if (in_port->active_iface != in_iface->port_ifidx) {
2247 /* Drop all multicast packets on inactive slaves. */
2252 /* Drop all packets for which we have learned a different input
2253 * port, because we probably sent the packet on one slave and got
2254 * it back on the other. Gratuitous ARP packets are an exception
2255 * to this rule: the host has moved to another switch. The exception
2256 * to the exception is if we locked the learning table to avoid
2257 * reflections on bond slaves. If this is the case, just drop the
2259 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2260 &is_grat_arp_locked);
2261 if (src_idx != -1 && src_idx != in_port->port_idx &&
2262 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2270 /* If the composed actions may be applied to any packet in the given 'flow',
2271 * returns true. Otherwise, the actions should only be applied to 'packet', or
2272 * not at all, if 'packet' was NULL. */
2274 process_flow(struct bridge *br, const flow_t *flow,
2275 const struct ofpbuf *packet, struct odp_actions *actions,
2276 tag_type *tags, uint16_t *nf_output_iface)
2278 struct port *in_port;
2279 struct port *out_port;
2283 /* Check whether we should drop packets in this flow. */
2284 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2289 /* Learn source MAC (but don't try to learn from revalidation). */
2291 update_learning_table(br, flow, vlan, in_port);
2294 /* Determine output port. */
2295 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2297 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2298 out_port = br->ports[out_port_idx];
2299 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2300 /* If we are revalidating but don't have a learning entry then
2301 * eject the flow. Installing a flow that floods packets opens
2302 * up a window of time where we could learn from a packet reflected
2303 * on a bond and blackhole packets before the learning table is
2304 * updated to reflect the correct port. */
2307 out_port = FLOOD_PORT;
2310 /* Don't send packets out their input ports. */
2311 if (in_port == out_port) {
2317 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2324 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2327 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2328 const struct ofp_phy_port *opp,
2331 struct bridge *br = br_;
2332 struct iface *iface;
2335 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2341 if (reason == OFPPR_DELETE) {
2342 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2343 br->name, iface->name);
2344 iface_destroy(iface);
2345 if (!port->n_ifaces) {
2346 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2347 br->name, port->name);
2353 if (port->n_ifaces > 1) {
2354 bool up = !(opp->state & OFPPS_LINK_DOWN);
2355 bond_link_status_update(iface, up);
2356 port_update_bond_compat(port);
2362 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2363 struct odp_actions *actions, tag_type *tags,
2364 uint16_t *nf_output_iface, void *br_)
2366 struct bridge *br = br_;
2368 COVERAGE_INC(bridge_process_flow);
2369 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2373 bridge_account_flow_ofhook_cb(const flow_t *flow,
2374 const union odp_action *actions,
2375 size_t n_actions, unsigned long long int n_bytes,
2378 struct bridge *br = br_;
2379 const union odp_action *a;
2380 struct port *in_port;
2384 /* Feed information from the active flows back into the learning table
2385 * to ensure that table is always in sync with what is actually flowing
2386 * through the datapath. */
2387 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2388 update_learning_table(br, flow, vlan, in_port);
2391 if (!br->has_bonded_ports) {
2395 for (a = actions; a < &actions[n_actions]; a++) {
2396 if (a->type == ODPAT_OUTPUT) {
2397 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2398 if (out_port && out_port->n_ifaces >= 2) {
2399 struct bond_entry *e = lookup_bond_entry(out_port,
2401 e->tx_bytes += n_bytes;
2408 bridge_account_checkpoint_ofhook_cb(void *br_)
2410 struct bridge *br = br_;
2414 if (!br->has_bonded_ports) {
2419 for (i = 0; i < br->n_ports; i++) {
2420 struct port *port = br->ports[i];
2421 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2422 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2423 bond_rebalance_port(port);
2428 static struct ofhooks bridge_ofhooks = {
2429 bridge_port_changed_ofhook_cb,
2430 bridge_normal_ofhook_cb,
2431 bridge_account_flow_ofhook_cb,
2432 bridge_account_checkpoint_ofhook_cb,
2435 /* Bonding functions. */
2437 /* Statistics for a single interface on a bonded port, used for load-based
2438 * bond rebalancing. */
2439 struct slave_balance {
2440 struct iface *iface; /* The interface. */
2441 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2443 /* All the "bond_entry"s that are assigned to this interface, in order of
2444 * increasing tx_bytes. */
2445 struct bond_entry **hashes;
2449 /* Sorts pointers to pointers to bond_entries in ascending order by the
2450 * interface to which they are assigned, and within a single interface in
2451 * ascending order of bytes transmitted. */
2453 compare_bond_entries(const void *a_, const void *b_)
2455 const struct bond_entry *const *ap = a_;
2456 const struct bond_entry *const *bp = b_;
2457 const struct bond_entry *a = *ap;
2458 const struct bond_entry *b = *bp;
2459 if (a->iface_idx != b->iface_idx) {
2460 return a->iface_idx > b->iface_idx ? 1 : -1;
2461 } else if (a->tx_bytes != b->tx_bytes) {
2462 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2468 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2469 * *descending* order by number of bytes transmitted. */
2471 compare_slave_balance(const void *a_, const void *b_)
2473 const struct slave_balance *a = a_;
2474 const struct slave_balance *b = b_;
2475 if (a->iface->enabled != b->iface->enabled) {
2476 return a->iface->enabled ? -1 : 1;
2477 } else if (a->tx_bytes != b->tx_bytes) {
2478 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2485 swap_bals(struct slave_balance *a, struct slave_balance *b)
2487 struct slave_balance tmp = *a;
2492 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2493 * given that 'p' (and only 'p') might be in the wrong location.
2495 * This function invalidates 'p', since it might now be in a different memory
2498 resort_bals(struct slave_balance *p,
2499 struct slave_balance bals[], size_t n_bals)
2502 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2503 swap_bals(p, p - 1);
2505 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2506 swap_bals(p, p + 1);
2512 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2514 if (VLOG_IS_DBG_ENABLED()) {
2515 struct ds ds = DS_EMPTY_INITIALIZER;
2516 const struct slave_balance *b;
2518 for (b = bals; b < bals + n_bals; b++) {
2522 ds_put_char(&ds, ',');
2524 ds_put_format(&ds, " %s %"PRIu64"kB",
2525 b->iface->name, b->tx_bytes / 1024);
2527 if (!b->iface->enabled) {
2528 ds_put_cstr(&ds, " (disabled)");
2530 if (b->n_hashes > 0) {
2531 ds_put_cstr(&ds, " (");
2532 for (i = 0; i < b->n_hashes; i++) {
2533 const struct bond_entry *e = b->hashes[i];
2535 ds_put_cstr(&ds, " + ");
2537 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2538 e - port->bond_hash, e->tx_bytes / 1024);
2540 ds_put_cstr(&ds, ")");
2543 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2548 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2550 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2553 struct bond_entry *hash = from->hashes[hash_idx];
2554 struct port *port = from->iface->port;
2555 uint64_t delta = hash->tx_bytes;
2557 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2558 "from %s to %s (now carrying %"PRIu64"kB and "
2559 "%"PRIu64"kB load, respectively)",
2560 port->name, delta / 1024, hash - port->bond_hash,
2561 from->iface->name, to->iface->name,
2562 (from->tx_bytes - delta) / 1024,
2563 (to->tx_bytes + delta) / 1024);
2565 /* Delete element from from->hashes.
2567 * We don't bother to add the element to to->hashes because not only would
2568 * it require more work, the only purpose it would be to allow that hash to
2569 * be migrated to another slave in this rebalancing run, and there is no
2570 * point in doing that. */
2571 if (hash_idx == 0) {
2574 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2575 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2579 /* Shift load away from 'from' to 'to'. */
2580 from->tx_bytes -= delta;
2581 to->tx_bytes += delta;
2583 /* Arrange for flows to be revalidated. */
2584 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2585 hash->iface_idx = to->iface->port_ifidx;
2586 hash->iface_tag = tag_create_random();
2590 bond_rebalance_port(struct port *port)
2592 struct slave_balance bals[DP_MAX_PORTS];
2594 struct bond_entry *hashes[BOND_MASK + 1];
2595 struct slave_balance *b, *from, *to;
2596 struct bond_entry *e;
2599 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2600 * descending order of tx_bytes, so that bals[0] represents the most
2601 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2604 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2605 * array for each slave_balance structure, we sort our local array of
2606 * hashes in order by slave, so that all of the hashes for a given slave
2607 * become contiguous in memory, and then we point each 'hashes' members of
2608 * a slave_balance structure to the start of a contiguous group. */
2609 n_bals = port->n_ifaces;
2610 for (b = bals; b < &bals[n_bals]; b++) {
2611 b->iface = port->ifaces[b - bals];
2616 for (i = 0; i <= BOND_MASK; i++) {
2617 hashes[i] = &port->bond_hash[i];
2619 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2620 for (i = 0; i <= BOND_MASK; i++) {
2622 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2623 b = &bals[e->iface_idx];
2624 b->tx_bytes += e->tx_bytes;
2626 b->hashes = &hashes[i];
2631 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2632 log_bals(bals, n_bals, port);
2634 /* Discard slaves that aren't enabled (which were sorted to the back of the
2635 * array earlier). */
2636 while (!bals[n_bals - 1].iface->enabled) {
2643 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2644 to = &bals[n_bals - 1];
2645 for (from = bals; from < to; ) {
2646 uint64_t overload = from->tx_bytes - to->tx_bytes;
2647 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2648 /* The extra load on 'from' (and all less-loaded slaves), compared
2649 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2650 * it is less than ~1Mbps. No point in rebalancing. */
2652 } else if (from->n_hashes == 1) {
2653 /* 'from' only carries a single MAC hash, so we can't shift any
2654 * load away from it, even though we want to. */
2657 /* 'from' is carrying significantly more load than 'to', and that
2658 * load is split across at least two different hashes. Pick a hash
2659 * to migrate to 'to' (the least-loaded slave), given that doing so
2660 * must decrease the ratio of the load on the two slaves by at
2663 * The sort order we use means that we prefer to shift away the
2664 * smallest hashes instead of the biggest ones. There is little
2665 * reason behind this decision; we could use the opposite sort
2666 * order to shift away big hashes ahead of small ones. */
2670 for (i = 0; i < from->n_hashes; i++) {
2671 double old_ratio, new_ratio;
2672 uint64_t delta = from->hashes[i]->tx_bytes;
2674 if (delta == 0 || from->tx_bytes - delta == 0) {
2675 /* Pointless move. */
2679 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2681 if (to->tx_bytes == 0) {
2682 /* Nothing on the new slave, move it. */
2686 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2687 new_ratio = (double)(from->tx_bytes - delta) /
2688 (to->tx_bytes + delta);
2690 if (new_ratio == 0) {
2691 /* Should already be covered but check to prevent division
2696 if (new_ratio < 1) {
2697 new_ratio = 1 / new_ratio;
2700 if (old_ratio - new_ratio > 0.1) {
2701 /* Would decrease the ratio, move it. */
2705 if (i < from->n_hashes) {
2706 bond_shift_load(from, to, i);
2707 port->bond_compat_is_stale = true;
2709 /* If the result of the migration changed the relative order of
2710 * 'from' and 'to' swap them back to maintain invariants. */
2711 if (order_swapped) {
2712 swap_bals(from, to);
2715 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2716 * point to different slave_balance structures. It is only
2717 * valid to do these two operations in a row at all because we
2718 * know that 'from' will not move past 'to' and vice versa. */
2719 resort_bals(from, bals, n_bals);
2720 resort_bals(to, bals, n_bals);
2727 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2728 * historical data to decay to <1% in 7 rebalancing runs. */
2729 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2735 bond_send_learning_packets(struct port *port)
2737 struct bridge *br = port->bridge;
2738 struct mac_entry *e;
2739 struct ofpbuf packet;
2740 int error, n_packets, n_errors;
2742 if (!port->n_ifaces || port->active_iface < 0) {
2746 ofpbuf_init(&packet, 128);
2747 error = n_packets = n_errors = 0;
2748 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2749 union ofp_action actions[2], *a;
2755 if (e->port == port->port_idx
2756 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2760 /* Compose actions. */
2761 memset(actions, 0, sizeof actions);
2764 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2765 a->vlan_vid.len = htons(sizeof *a);
2766 a->vlan_vid.vlan_vid = htons(e->vlan);
2769 a->output.type = htons(OFPAT_OUTPUT);
2770 a->output.len = htons(sizeof *a);
2771 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2776 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2778 flow_extract(&packet, 0, ODPP_NONE, &flow);
2779 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2786 ofpbuf_uninit(&packet);
2789 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2790 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2791 "packets, last error was: %s",
2792 port->name, n_errors, n_packets, strerror(error));
2794 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2795 port->name, n_packets);
2799 /* Bonding unixctl user interface functions. */
2802 bond_unixctl_list(struct unixctl_conn *conn,
2803 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2805 struct ds ds = DS_EMPTY_INITIALIZER;
2806 const struct bridge *br;
2808 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2810 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2813 for (i = 0; i < br->n_ports; i++) {
2814 const struct port *port = br->ports[i];
2815 if (port->n_ifaces > 1) {
2818 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2819 for (j = 0; j < port->n_ifaces; j++) {
2820 const struct iface *iface = port->ifaces[j];
2822 ds_put_cstr(&ds, ", ");
2824 ds_put_cstr(&ds, iface->name);
2826 ds_put_char(&ds, '\n');
2830 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2834 static struct port *
2835 bond_find(const char *name)
2837 const struct bridge *br;
2839 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2842 for (i = 0; i < br->n_ports; i++) {
2843 struct port *port = br->ports[i];
2844 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2853 bond_unixctl_show(struct unixctl_conn *conn,
2854 const char *args, void *aux OVS_UNUSED)
2856 struct ds ds = DS_EMPTY_INITIALIZER;
2857 const struct port *port;
2860 port = bond_find(args);
2862 unixctl_command_reply(conn, 501, "no such bond");
2866 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2867 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2868 ds_put_format(&ds, "next rebalance: %lld ms\n",
2869 port->bond_next_rebalance - time_msec());
2870 for (j = 0; j < port->n_ifaces; j++) {
2871 const struct iface *iface = port->ifaces[j];
2872 struct bond_entry *be;
2875 ds_put_format(&ds, "slave %s: %s\n",
2876 iface->name, iface->enabled ? "enabled" : "disabled");
2877 if (j == port->active_iface) {
2878 ds_put_cstr(&ds, "\tactive slave\n");
2880 if (iface->delay_expires != LLONG_MAX) {
2881 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2882 iface->enabled ? "downdelay" : "updelay",
2883 iface->delay_expires - time_msec());
2887 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2888 int hash = be - port->bond_hash;
2889 struct mac_entry *me;
2891 if (be->iface_idx != j) {
2895 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2896 hash, be->tx_bytes / 1024);
2899 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2900 &port->bridge->ml->lrus) {
2903 if (bond_hash(me->mac) == hash
2904 && me->port != port->port_idx
2905 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2906 && dp_ifidx == iface->dp_ifidx)
2908 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2909 ETH_ADDR_ARGS(me->mac));
2914 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2919 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2920 void *aux OVS_UNUSED)
2922 char *args = (char *) args_;
2923 char *save_ptr = NULL;
2924 char *bond_s, *hash_s, *slave_s;
2925 uint8_t mac[ETH_ADDR_LEN];
2927 struct iface *iface;
2928 struct bond_entry *entry;
2931 bond_s = strtok_r(args, " ", &save_ptr);
2932 hash_s = strtok_r(NULL, " ", &save_ptr);
2933 slave_s = strtok_r(NULL, " ", &save_ptr);
2935 unixctl_command_reply(conn, 501,
2936 "usage: bond/migrate BOND HASH SLAVE");
2940 port = bond_find(bond_s);
2942 unixctl_command_reply(conn, 501, "no such bond");
2946 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2947 == ETH_ADDR_SCAN_COUNT) {
2948 hash = bond_hash(mac);
2949 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2950 hash = atoi(hash_s) & BOND_MASK;
2952 unixctl_command_reply(conn, 501, "bad hash");
2956 iface = port_lookup_iface(port, slave_s);
2958 unixctl_command_reply(conn, 501, "no such slave");
2962 if (!iface->enabled) {
2963 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2967 entry = &port->bond_hash[hash];
2968 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2969 entry->iface_idx = iface->port_ifidx;
2970 entry->iface_tag = tag_create_random();
2971 port->bond_compat_is_stale = true;
2972 unixctl_command_reply(conn, 200, "migrated");
2976 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2977 void *aux OVS_UNUSED)
2979 char *args = (char *) args_;
2980 char *save_ptr = NULL;
2981 char *bond_s, *slave_s;
2983 struct iface *iface;
2985 bond_s = strtok_r(args, " ", &save_ptr);
2986 slave_s = strtok_r(NULL, " ", &save_ptr);
2988 unixctl_command_reply(conn, 501,
2989 "usage: bond/set-active-slave BOND SLAVE");
2993 port = bond_find(bond_s);
2995 unixctl_command_reply(conn, 501, "no such bond");
2999 iface = port_lookup_iface(port, slave_s);
3001 unixctl_command_reply(conn, 501, "no such slave");
3005 if (!iface->enabled) {
3006 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3010 if (port->active_iface != iface->port_ifidx) {
3011 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3012 port->active_iface = iface->port_ifidx;
3013 port->active_iface_tag = tag_create_random();
3014 VLOG_INFO("port %s: active interface is now %s",
3015 port->name, iface->name);
3016 bond_send_learning_packets(port);
3017 unixctl_command_reply(conn, 200, "done");
3019 unixctl_command_reply(conn, 200, "no change");
3024 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3026 char *args = (char *) args_;
3027 char *save_ptr = NULL;
3028 char *bond_s, *slave_s;
3030 struct iface *iface;
3032 bond_s = strtok_r(args, " ", &save_ptr);
3033 slave_s = strtok_r(NULL, " ", &save_ptr);
3035 unixctl_command_reply(conn, 501,
3036 "usage: bond/enable/disable-slave BOND SLAVE");
3040 port = bond_find(bond_s);
3042 unixctl_command_reply(conn, 501, "no such bond");
3046 iface = port_lookup_iface(port, slave_s);
3048 unixctl_command_reply(conn, 501, "no such slave");
3052 bond_enable_slave(iface, enable);
3053 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3057 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3058 void *aux OVS_UNUSED)
3060 enable_slave(conn, args, true);
3064 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3065 void *aux OVS_UNUSED)
3067 enable_slave(conn, args, false);
3071 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3072 void *aux OVS_UNUSED)
3074 uint8_t mac[ETH_ADDR_LEN];
3078 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3079 == ETH_ADDR_SCAN_COUNT) {
3080 hash = bond_hash(mac);
3082 hash_cstr = xasprintf("%u", hash);
3083 unixctl_command_reply(conn, 200, hash_cstr);
3086 unixctl_command_reply(conn, 501, "invalid mac");
3093 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3094 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3095 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3096 unixctl_command_register("bond/set-active-slave",
3097 bond_unixctl_set_active_slave, NULL);
3098 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3100 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3102 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3105 /* Port functions. */
3107 static struct port *
3108 port_create(struct bridge *br, const char *name)
3112 port = xzalloc(sizeof *port);
3114 port->port_idx = br->n_ports;
3116 port->trunks = NULL;
3117 port->name = xstrdup(name);
3118 port->active_iface = -1;
3120 if (br->n_ports >= br->allocated_ports) {
3121 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3124 br->ports[br->n_ports++] = port;
3125 shash_add_assert(&br->port_by_name, port->name, port);
3127 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3134 get_port_other_config(const struct ovsrec_port *port, const char *key,
3135 const char *default_value)
3137 const char *value = get_ovsrec_key_value(key,
3138 port->key_other_config,
3139 port->value_other_config,
3140 port->n_other_config);
3141 return value ? value : default_value;
3145 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3147 struct shash new_ifaces;
3150 /* Collect list of new interfaces. */
3151 shash_init(&new_ifaces);
3152 for (i = 0; i < cfg->n_interfaces; i++) {
3153 const char *name = cfg->interfaces[i]->name;
3154 shash_add_once(&new_ifaces, name, NULL);
3157 /* Get rid of deleted interfaces. */
3158 for (i = 0; i < port->n_ifaces; ) {
3159 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3160 iface_destroy(port->ifaces[i]);
3166 shash_destroy(&new_ifaces);
3170 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3172 struct shash new_ifaces;
3173 long long int next_rebalance;
3174 unsigned long *trunks;
3180 /* Update settings. */
3181 port->updelay = cfg->bond_updelay;
3182 if (port->updelay < 0) {
3185 port->updelay = cfg->bond_downdelay;
3186 if (port->downdelay < 0) {
3187 port->downdelay = 0;
3189 port->bond_rebalance_interval = atoi(
3190 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3191 if (port->bond_rebalance_interval < 1000) {
3192 port->bond_rebalance_interval = 1000;
3194 next_rebalance = time_msec() + port->bond_rebalance_interval;
3195 if (port->bond_next_rebalance > next_rebalance) {
3196 port->bond_next_rebalance = next_rebalance;
3199 /* Add new interfaces and update 'cfg' member of existing ones. */
3200 shash_init(&new_ifaces);
3201 for (i = 0; i < cfg->n_interfaces; i++) {
3202 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3203 struct iface *iface;
3205 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3206 VLOG_WARN("port %s: %s specified twice as port interface",
3207 port->name, if_cfg->name);
3211 iface = iface_lookup(port->bridge, if_cfg->name);
3213 if (iface->port != port) {
3214 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3216 port->bridge->name, if_cfg->name, iface->port->name);
3219 iface->cfg = if_cfg;
3221 iface_create(port, if_cfg);
3224 shash_destroy(&new_ifaces);
3229 if (port->n_ifaces < 2) {
3231 if (vlan >= 0 && vlan <= 4095) {
3232 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3237 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3238 * they even work as-is. But they have not been tested. */
3239 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3243 if (port->vlan != vlan) {
3245 bridge_flush(port->bridge);
3248 /* Get trunked VLANs. */
3250 if (vlan < 0 && cfg->n_trunks) {
3254 trunks = bitmap_allocate(4096);
3256 for (i = 0; i < cfg->n_trunks; i++) {
3257 int trunk = cfg->trunks[i];
3259 bitmap_set1(trunks, trunk);
3265 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3266 port->name, cfg->n_trunks);
3268 if (n_errors == cfg->n_trunks) {
3269 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3271 bitmap_free(trunks);
3274 } else if (vlan >= 0 && cfg->n_trunks) {
3275 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3279 ? port->trunks != NULL
3280 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3281 bridge_flush(port->bridge);
3283 bitmap_free(port->trunks);
3284 port->trunks = trunks;
3288 port_destroy(struct port *port)
3291 struct bridge *br = port->bridge;
3295 proc_net_compat_update_vlan(port->name, NULL, 0);
3296 proc_net_compat_update_bond(port->name, NULL);
3298 for (i = 0; i < MAX_MIRRORS; i++) {
3299 struct mirror *m = br->mirrors[i];
3300 if (m && m->out_port == port) {
3305 while (port->n_ifaces > 0) {
3306 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3309 shash_find_and_delete_assert(&br->port_by_name, port->name);
3311 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3312 del->port_idx = port->port_idx;
3315 bitmap_free(port->trunks);
3322 static struct port *
3323 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3325 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3326 return iface ? iface->port : NULL;
3329 static struct port *
3330 port_lookup(const struct bridge *br, const char *name)
3332 return shash_find_data(&br->port_by_name, name);
3335 static struct iface *
3336 port_lookup_iface(const struct port *port, const char *name)
3338 struct iface *iface = iface_lookup(port->bridge, name);
3339 return iface && iface->port == port ? iface : NULL;
3343 port_update_bonding(struct port *port)
3345 if (port->n_ifaces < 2) {
3346 /* Not a bonded port. */
3347 if (port->bond_hash) {
3348 free(port->bond_hash);
3349 port->bond_hash = NULL;
3350 port->bond_compat_is_stale = true;
3351 port->bond_fake_iface = false;
3354 if (!port->bond_hash) {
3357 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3358 for (i = 0; i <= BOND_MASK; i++) {
3359 struct bond_entry *e = &port->bond_hash[i];
3363 port->no_ifaces_tag = tag_create_random();
3364 bond_choose_active_iface(port);
3365 port->bond_next_rebalance
3366 = time_msec() + port->bond_rebalance_interval;
3368 if (port->cfg->bond_fake_iface) {
3369 port->bond_next_fake_iface_update = time_msec();
3372 port->bond_compat_is_stale = true;
3373 port->bond_fake_iface = port->cfg->bond_fake_iface;
3378 port_update_bond_compat(struct port *port)
3380 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3381 struct compat_bond bond;
3384 if (port->n_ifaces < 2) {
3385 proc_net_compat_update_bond(port->name, NULL);
3390 bond.updelay = port->updelay;
3391 bond.downdelay = port->downdelay;
3394 bond.hashes = compat_hashes;
3395 if (port->bond_hash) {
3396 const struct bond_entry *e;
3397 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3398 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3399 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3400 cbh->hash = e - port->bond_hash;
3401 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3406 bond.n_slaves = port->n_ifaces;
3407 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3408 for (i = 0; i < port->n_ifaces; i++) {
3409 struct iface *iface = port->ifaces[i];
3410 struct compat_bond_slave *slave = &bond.slaves[i];
3411 slave->name = iface->name;
3413 /* We need to make the same determination as the Linux bonding
3414 * code to determine whether a slave should be consider "up".
3415 * The Linux function bond_miimon_inspect() supports four
3416 * BOND_LINK_* states:
3418 * - BOND_LINK_UP: carrier detected, updelay has passed.
3419 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3420 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3421 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3423 * The function bond_info_show_slave() only considers BOND_LINK_UP
3424 * to be "up" and anything else to be "down".
3426 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3430 netdev_get_etheraddr(iface->netdev, slave->mac);
3433 if (port->bond_fake_iface) {
3434 struct netdev *bond_netdev;
3436 if (!netdev_open_default(port->name, &bond_netdev)) {
3438 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3440 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3442 netdev_close(bond_netdev);
3446 proc_net_compat_update_bond(port->name, &bond);
3451 port_update_vlan_compat(struct port *port)
3453 struct bridge *br = port->bridge;
3454 char *vlandev_name = NULL;
3456 if (port->vlan > 0) {
3457 /* Figure out the name that the VLAN device should actually have, if it
3458 * existed. This takes some work because the VLAN device would not
3459 * have port->name in its name; rather, it would have the trunk port's
3460 * name, and 'port' would be attached to a bridge that also had the
3461 * VLAN device one of its ports. So we need to find a trunk port that
3462 * includes port->vlan.
3464 * There might be more than one candidate. This doesn't happen on
3465 * XenServer, so if it happens we just pick the first choice in
3466 * alphabetical order instead of creating multiple VLAN devices. */
3468 for (i = 0; i < br->n_ports; i++) {
3469 struct port *p = br->ports[i];
3470 if (port_trunks_vlan(p, port->vlan)
3472 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3474 uint8_t ea[ETH_ADDR_LEN];
3475 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3476 if (!eth_addr_is_multicast(ea) &&
3477 !eth_addr_is_reserved(ea) &&
3478 !eth_addr_is_zero(ea)) {
3479 vlandev_name = p->name;
3484 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3487 /* Interface functions. */
3489 static struct iface *
3490 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3492 struct bridge *br = port->bridge;
3493 struct iface *iface;
3494 char *name = if_cfg->name;
3497 iface = xzalloc(sizeof *iface);
3499 iface->port_ifidx = port->n_ifaces;
3500 iface->name = xstrdup(name);
3501 iface->dp_ifidx = -1;
3502 iface->tag = tag_create_random();
3503 iface->delay_expires = LLONG_MAX;
3504 iface->netdev = NULL;
3505 iface->cfg = if_cfg;
3507 shash_add_assert(&br->iface_by_name, iface->name, iface);
3509 /* Attempt to create the network interface in case it doesn't exist yet. */
3510 if (!iface_is_internal(br, iface->name)) {
3511 error = set_up_iface(if_cfg, iface, true);
3513 VLOG_WARN("could not create iface %s: %s", iface->name,
3516 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3523 if (port->n_ifaces >= port->allocated_ifaces) {
3524 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3525 sizeof *port->ifaces);
3527 port->ifaces[port->n_ifaces++] = iface;
3528 if (port->n_ifaces > 1) {
3529 br->has_bonded_ports = true;
3532 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3540 iface_destroy(struct iface *iface)
3543 struct port *port = iface->port;
3544 struct bridge *br = port->bridge;
3545 bool del_active = port->active_iface == iface->port_ifidx;
3548 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3550 if (iface->dp_ifidx >= 0) {
3551 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3554 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3555 del->port_ifidx = iface->port_ifidx;
3557 netdev_close(iface->netdev);
3560 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3561 bond_choose_active_iface(port);
3562 bond_send_learning_packets(port);
3568 bridge_flush(port->bridge);
3572 static struct iface *
3573 iface_lookup(const struct bridge *br, const char *name)
3575 return shash_find_data(&br->iface_by_name, name);
3578 static struct iface *
3579 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3581 return port_array_get(&br->ifaces, dp_ifidx);
3584 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3585 * 'br', that is, an interface that is entirely simulated within the datapath.
3586 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3587 * interfaces are created by setting "iface.<iface>.internal = true".
3589 * In addition, we have a kluge-y feature that creates an internal port with
3590 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3591 * This feature needs to go away in the long term. Until then, this is one
3592 * reason why this function takes a name instead of a struct iface: the fake
3593 * interfaces created this way do not have a struct iface. */
3595 iface_is_internal(const struct bridge *br, const char *if_name)
3597 struct iface *iface;
3600 if (!strcmp(if_name, br->name)) {
3604 iface = iface_lookup(br, if_name);
3605 if (iface && !strcmp(iface->cfg->type, "internal")) {
3609 port = port_lookup(br, if_name);
3610 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3616 /* Set Ethernet address of 'iface', if one is specified in the configuration
3619 iface_set_mac(struct iface *iface)
3621 uint8_t ea[ETH_ADDR_LEN];
3623 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3624 if (eth_addr_is_multicast(ea)) {
3625 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3627 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3628 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3629 iface->name, iface->name);
3631 int error = netdev_set_etheraddr(iface->netdev, ea);
3633 VLOG_ERR("interface %s: setting MAC failed (%s)",
3634 iface->name, strerror(error));
3641 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3642 struct shash *shash)
3647 for (i = 0; i < n; i++) {
3648 shash_add(shash, keys[i], values[i]);
3652 struct iface_delete_queues_cbdata {
3653 struct netdev *netdev;
3654 const int64_t *queue_ids;
3659 queue_ids_include(const int64_t *ids, size_t n, int64_t target)
3664 while (low < high) {
3665 size_t mid = low + (high - low) / 2;
3666 if (target > ids[mid]) {
3668 } else if (target < ids[mid]) {
3678 iface_delete_queues(unsigned int queue_id,
3679 const struct shash *details OVS_UNUSED, void *cbdata_)
3681 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3683 if (!queue_ids_include(cbdata->queue_ids, cbdata->n_queue_ids, queue_id)) {
3684 netdev_delete_queue(cbdata->netdev, queue_id);
3689 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3691 if (!qos || qos->type[0] == '\0') {
3692 netdev_set_qos(iface->netdev, NULL, NULL);
3694 struct iface_delete_queues_cbdata cbdata;
3695 struct shash details;
3698 /* Configure top-level Qos for 'iface'. */
3699 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3700 qos->n_other_config, &details);
3701 netdev_set_qos(iface->netdev, qos->type, &details);
3702 shash_destroy(&details);
3704 /* Deconfigure queues that were deleted. */
3705 cbdata.netdev = iface->netdev;
3706 cbdata.queue_ids = qos->key_queues;
3707 cbdata.n_queue_ids = qos->n_queues;
3708 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3710 /* Configure queues for 'iface'. */
3711 for (i = 0; i < qos->n_queues; i++) {
3712 const struct ovsrec_queue *queue = qos->value_queues[i];
3713 unsigned int queue_id = qos->key_queues[i];
3715 shash_from_ovs_idl_map(queue->key_other_config,
3716 queue->value_other_config,
3717 queue->n_other_config, &details);
3718 netdev_set_queue(iface->netdev, queue_id, &details);
3719 shash_destroy(&details);
3724 /* Port mirroring. */
3727 mirror_reconfigure(struct bridge *br)
3729 struct shash old_mirrors, new_mirrors;
3730 struct shash_node *node;
3731 unsigned long *rspan_vlans;
3734 /* Collect old mirrors. */
3735 shash_init(&old_mirrors);
3736 for (i = 0; i < MAX_MIRRORS; i++) {
3737 if (br->mirrors[i]) {
3738 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3742 /* Collect new mirrors. */
3743 shash_init(&new_mirrors);
3744 for (i = 0; i < br->cfg->n_mirrors; i++) {
3745 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3746 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3747 VLOG_WARN("bridge %s: %s specified twice as mirror",
3748 br->name, cfg->name);
3752 /* Get rid of deleted mirrors and add new mirrors. */
3753 SHASH_FOR_EACH (node, &old_mirrors) {
3754 if (!shash_find(&new_mirrors, node->name)) {
3755 mirror_destroy(node->data);
3758 SHASH_FOR_EACH (node, &new_mirrors) {
3759 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3761 mirror = mirror_create(br, node->name);
3766 mirror_reconfigure_one(mirror, node->data);
3768 shash_destroy(&old_mirrors);
3769 shash_destroy(&new_mirrors);
3771 /* Update port reserved status. */
3772 for (i = 0; i < br->n_ports; i++) {
3773 br->ports[i]->is_mirror_output_port = false;
3775 for (i = 0; i < MAX_MIRRORS; i++) {
3776 struct mirror *m = br->mirrors[i];
3777 if (m && m->out_port) {
3778 m->out_port->is_mirror_output_port = true;
3782 /* Update flooded vlans (for RSPAN). */
3784 if (br->cfg->n_flood_vlans) {
3785 rspan_vlans = bitmap_allocate(4096);
3787 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3788 int64_t vlan = br->cfg->flood_vlans[i];
3789 if (vlan >= 0 && vlan < 4096) {
3790 bitmap_set1(rspan_vlans, vlan);
3791 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3794 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3799 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3804 static struct mirror *
3805 mirror_create(struct bridge *br, const char *name)
3810 for (i = 0; ; i++) {
3811 if (i >= MAX_MIRRORS) {
3812 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3813 "cannot create %s", br->name, MAX_MIRRORS, name);
3816 if (!br->mirrors[i]) {
3821 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3824 br->mirrors[i] = m = xzalloc(sizeof *m);
3827 m->name = xstrdup(name);
3828 shash_init(&m->src_ports);
3829 shash_init(&m->dst_ports);
3839 mirror_destroy(struct mirror *m)
3842 struct bridge *br = m->bridge;
3845 for (i = 0; i < br->n_ports; i++) {
3846 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3847 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3850 shash_destroy(&m->src_ports);
3851 shash_destroy(&m->dst_ports);
3854 m->bridge->mirrors[m->idx] = NULL;
3862 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3863 struct shash *names)
3867 for (i = 0; i < n_ports; i++) {
3868 const char *name = ports[i]->name;
3869 if (port_lookup(m->bridge, name)) {
3870 shash_add_once(names, name, NULL);
3872 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3873 "port %s", m->bridge->name, m->name, name);
3879 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3885 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3887 for (i = 0; i < cfg->n_select_vlan; i++) {
3888 int64_t vlan = cfg->select_vlan[i];
3889 if (vlan < 0 || vlan > 4095) {
3890 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3891 m->bridge->name, m->name, vlan);
3893 (*vlans)[n_vlans++] = vlan;
3900 vlan_is_mirrored(const struct mirror *m, int vlan)
3904 for (i = 0; i < m->n_vlans; i++) {
3905 if (m->vlans[i] == vlan) {
3913 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3917 for (i = 0; i < m->n_vlans; i++) {
3918 if (port_trunks_vlan(p, m->vlans[i])) {
3926 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3928 struct shash src_ports, dst_ports;
3929 mirror_mask_t mirror_bit;
3930 struct port *out_port;
3936 /* Get output port. */
3937 if (cfg->output_port) {
3938 out_port = port_lookup(m->bridge, cfg->output_port->name);
3940 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3941 m->bridge->name, m->name);
3947 if (cfg->output_vlan) {
3948 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3949 "output vlan; ignoring output vlan",
3950 m->bridge->name, m->name);
3952 } else if (cfg->output_vlan) {
3954 out_vlan = *cfg->output_vlan;
3956 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3957 m->bridge->name, m->name);
3962 shash_init(&src_ports);
3963 shash_init(&dst_ports);
3964 if (cfg->select_all) {
3965 for (i = 0; i < m->bridge->n_ports; i++) {
3966 const char *name = m->bridge->ports[i]->name;
3967 shash_add_once(&src_ports, name, NULL);
3968 shash_add_once(&dst_ports, name, NULL);
3973 /* Get ports, and drop duplicates and ports that don't exist. */
3974 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3976 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3979 /* Get all the vlans, and drop duplicate and invalid vlans. */
3980 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3983 /* Update mirror data. */
3984 if (!shash_equal_keys(&m->src_ports, &src_ports)
3985 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3986 || m->n_vlans != n_vlans
3987 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3988 || m->out_port != out_port
3989 || m->out_vlan != out_vlan) {
3990 bridge_flush(m->bridge);
3992 shash_swap(&m->src_ports, &src_ports);
3993 shash_swap(&m->dst_ports, &dst_ports);
3996 m->n_vlans = n_vlans;
3997 m->out_port = out_port;
3998 m->out_vlan = out_vlan;
4001 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4002 for (i = 0; i < m->bridge->n_ports; i++) {
4003 struct port *port = m->bridge->ports[i];
4005 if (shash_find(&m->src_ports, port->name)
4008 ? port_trunks_any_mirrored_vlan(m, port)
4009 : vlan_is_mirrored(m, port->vlan)))) {
4010 port->src_mirrors |= mirror_bit;
4012 port->src_mirrors &= ~mirror_bit;
4015 if (shash_find(&m->dst_ports, port->name)) {
4016 port->dst_mirrors |= mirror_bit;
4018 port->dst_mirrors &= ~mirror_bit;
4023 shash_destroy(&src_ports);
4024 shash_destroy(&dst_ports);